Alanjawi, Ali
(2009)
USING COEVOLUTION IN COMPLEX DOMAINS.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
Abstract
Genetic Algorithms is a computational model inspired by Darwin's theory of evolution. It has a broad range of applications from function optimization to solving robotic control problems. Coevolution is an extension of Genetic Algorithms in which more than one population is evolved at the same time. Coevolution can be done in two ways: cooperatively, in which populations jointly try to solve an evolutionary problem, or competitively. Coevolution has been shown to be useful in solving many problems, yet its application in complex domains still needs to be demonstrated.Robotic soccer is a complex domain that has a dynamic and noisy environment. Many Reinforcement Learning techniques have been applied to the robotic soccer domain, since it is a great test bed for many machine learning methods. However, the success of Reinforcement Learning methods has been limited due to the huge state space of the domain. Evolutionary Algorithms have also been used to tackle this domain; nevertheless, their application has been limited to a small subset of the domain, and no attempt has been shown to be successful in acting on solving the whole problem.This thesis will try to answer the question of whether coevolution can be applied successfully to complex domains. Three techniques are introduced to tackle the robotic soccer problem. First, an incremental learning algorithm is used to achieve a desirable performance of some soccer tasks. Second, a hierarchical coevolution paradigm is introduced to allow coevolution to scale up in solving the problem. Third, an orchestration mechanism is utilized to manage the learning processes.
Share
| Citation/Export: |
|
| Social Networking: |
|
Details
| Item Type: |
University of Pittsburgh ETD
|
| Status: |
Unpublished |
| Creators/Authors: |
|
| ETD Committee: |
|
| Date: |
30 September 2009 |
| Date Type: |
Completion |
| Defense Date: |
26 February 2009 |
| Approval Date: |
30 September 2009 |
| Submission Date: |
28 April 2009 |
| Access Restriction: |
5 year -- Restrict access to University of Pittsburgh for a period of 5 years. |
| Institution: |
University of Pittsburgh |
| Schools and Programs: |
Dietrich School of Arts and Sciences > Computer Science |
| Degree: |
PhD - Doctor of Philosophy |
| Thesis Type: |
Doctoral Dissertation |
| Refereed: |
Yes |
| Uncontrolled Keywords: |
Artificial intelligence; Coevolution; Evolutionary Algorithms; Macine Learning; Robotics |
| Other ID: |
http://etd.library.pitt.edu/ETD/available/etd-04282009-052315/, etd-04282009-052315 |
| Date Deposited: |
10 Nov 2011 19:42 |
| Last Modified: |
19 Dec 2016 14:35 |
| URI: |
http://d-scholarship.pitt.edu/id/eprint/7715 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[feed]](/images/feed-icon-32x32.png){kind=icon}